报告题目：Performance-Based Reinforcement Learning Control for 2-D Markov Jump Roesser Systems
报告人：吴佳成 博士
报告时间：2023年11月29日9点
报告地点：电气楼304
报告对象：感兴趣的教师、研究生等
主办单位：电气与十大网投平台信誉排行榜
报告人简介：
吴佳成，浙江大学控制科学与工程学院博士，已发表SCI论文10篇，其中IEEE汇刊6篇。目前研究兴趣包括，强化学习控制，Markov跳变系统，鲁棒控制等。
报告摘要：
This paper studies the performance-based reinforce- ment learning control problem for discrete-time two-dimensional Markov jump Roesser systems with unknown system dynamics. To stabilize two-dimensional Markov jump Roesser systems and minimize H∞ performance, existing design methods typically necessitate prior knowledge of the system dynamics. However, lack of model information is a common phenomenon in practice. To track this problem, a novel model-free reinforcement learning control method where horizontal and vertical system data are effectively utilized is presented. It is shown that the optimal control policy is designed with the minimized H∞ performance in the presence of worst-case disturbance. Moreover, a data- driven value iteration algorithm is developed for two-dimensional Markov jump Roesser systems, with the aim of searching for an initial stabilizing control policy. Compared with the existing control methods of two-dimensional Markov jump Roesser systems, the most significant advantage of the proposed method is that, by utilizing system data, optimal control policy and optimal H∞ performance can be obtained by solving a set of linear matrix inequalities depending on system data. Then, the convergence of the proposed algorithms and the asymptotic mean square stability of the closed-loop systems are analyzed. Finally, simulation results demonstrate the significance and validity of the proposed control methods.
欢迎全校师生踊跃参加！
